We describe Data Science as the practice to gain actionable insights from data and building smart applications for real-world problems. It's a deeply practical field that requires collaboration and the ability to iteratively improve a product. Full-stack data scientists are still rare in 2019.
2. What you’ll learn
● Three demos to warm up (5 mins)
● Discuss the term Data Science (5 mins)
● Object detection with Deep Learning (15 mins)
● Tackle a realistic use-case (10 mins)
● Summarize and discuss extensions (10 mins)
3. Demo - live at: https://data-science-demo.herokuapp.com/
4. Common ground
● Intuitive computer vision tasks
● Real-time
● Commoditized
● Impossible 5 years ago
● Use machine learning
● In fact: deep learning
5. Deep
Learning
Data Science is the
practice of gaining
actionable insights from data
and building smart products
for real-world problems
6. Cool applications, but...
● Black box, we didn’t build anything
● Actionable insights need clear goals
● At most part of a product
● Not at all real-world
Black box
Web app
7. What’s in the box?
On object detection
● Task: Find bounding boxes for objects in
images + correct labels
● Relatively easy for humans*
● You Only Look Once (YOLO) state of the art
detection algorithm
● Accurate and very fast (~30FPS)
● Other strong alternatives available
9. Lightning view on Deep Learning
Deep neural network
Dog
Dog
Input Predict
Learn
Compare (boxes & labels)
1 2
3
4
Update network
weights accordingly
Dog
Dog
Dog
Cat
Dog
Dog
11. Data Science Experimentation Flow
Data
Collection
Data
Cleaning
Data
Processing
Model
training
Model
definition
Model
evaluation
12. Use case: Security cameras
● Your company has 10.000 security cameras
● CEO wants gun* detection in real time (tomorrow)
● Dashboard for reporting & alerts
● You’re the lead data scientist on the project
14. Our Minimum Viable Product (MVP)
Data
collection
Data
processing
Model
training
AlgorithmWeb cam App
Reporting &
Alerting
1. Iteration
Data Science is the practice of
gaining actionable insights from
data and building smart products
for real-world problems
15. Parallel data processing & training
AlgorithmWeb cam App
Reporting &
Alerting
2. IterationProduction
DBs
Parallel
processing
17. Customer data & insights
App
Robust, low-latency, scales, secure
Reporting &
alerting
System
monitoring
Customer
dashboard
4. Iteration
18. Black Box vs. Complex System
App
Robust, low-latency, scales,
secure
Reporting
& alerting
System
monitoring
Customer
dashboard
Black box
Web app
19. Key Takeaways
● Data science is a complex, practical field - not just experimentation
● Iterative process & team effort
● Machine and deep learning often center stage
● Full-stack data scientists are still rare - your turn!